From 46f569f5711098f82cd141e878aa98261f90fa1a Mon Sep 17 00:00:00 2001
From: kijai <40791699+kijai@users.noreply.github.com>
Date: Wed, 18 Dec 2024 00:34:12 +0200
Subject: [PATCH] clearer error on frame count mismatch

---
 hyvideo/modules/models.py | 11 ++++++-----
 nodes.py                  |  4 +++-
 nodes_rf_inversion.py     |  6 +++---
 3 files changed, 12 insertions(+), 9 deletions(-)

diff --git a/hyvideo/modules/models.py b/hyvideo/modules/models.py
index a6f53f8..810f2b7 100644
--- a/hyvideo/modules/models.py
+++ b/hyvideo/modules/models.py
@@ -175,11 +175,12 @@ def forward(
 
         # Apply RoPE if needed.
         if freqs_cis is not None:
-            img_qq, img_kk = apply_rotary_emb(img_q, img_k, freqs_cis, head_first=False)
-            assert (
-                img_qq.shape == img_q.shape and img_kk.shape == img_k.shape
-            ), f"img_kk: {img_qq.shape}, img_q: {img_q.shape}, img_kk: {img_kk.shape}, img_k: {img_k.shape}"
-            img_q, img_k = img_qq, img_kk
+            img_q, img_k = apply_rotary_emb(img_q, img_k, freqs_cis, head_first=False)
+            #img_q, img_k = img_qq, img_kk
+            #assert (
+            #    img_qq.shape == img_q.shape and img_kk.shape == img_k.shape
+            #), f"img_kk: {img_qq.shape}, img_q: {img_q.shape}, img_kk: {img_kk.shape}, img_k: {img_k.shape}"
+            
 
         # Prepare txt for attention.
         txt_modulated = self.txt_norm1(txt)
diff --git a/nodes.py b/nodes.py
index efdabf6..b51e6e7 100644
--- a/nodes.py
+++ b/nodes.py
@@ -1087,7 +1087,7 @@ def process(self, model, hyvid_embeds, flow_shift, steps, embedded_guidance_scal
             )
         if (num_frames - 1) % 4 != 0:
             raise ValueError(
-                f"`video_length-1` must be a multiple of 4, got {num_frames}"
+                f"`video_length - 1 (that's minus one frame)` must be a multiple of 4, got {num_frames}"
             )
 
         log.info(
@@ -1101,6 +1101,8 @@ def process(self, model, hyvid_embeds, flow_shift, steps, embedded_guidance_scal
             transformer, num_frames, target_height, target_width
         )
         n_tokens = freqs_cos.shape[0]
+        freqs_cos = freqs_cos.to(dtype).to(device)
+        freqs_sin = freqs_sin.to(dtype).to(device)
 
        
 
diff --git a/nodes_rf_inversion.py b/nodes_rf_inversion.py
index 9879b97..4a6982b 100644
--- a/nodes_rf_inversion.py
+++ b/nodes_rf_inversion.py
@@ -110,7 +110,7 @@ def process(self, model, hyvid_embeds, flow_shift, steps, embedded_guidance_scal
             )
         if (num_frames - 1) % 4 != 0:
             raise ValueError(
-                f"`video_length-1` must be a multiple of 4, got {num_frames}"
+                 f"`video_length - 1 (that's minus one frame)` must be a multiple of 4, got {num_frames}"
             )
 
         log.info(
@@ -320,7 +320,7 @@ def process(self, model, hyvid_embeds, flow_shift, steps, embedded_guidance_scal
             )
         if (num_frames - 1) % 4 != 0:
             raise ValueError(
-                f"`video_length-1` must be a multiple of 4, got {num_frames}"
+                 f"`video_length - 1 (that's minus one frame)` must be a multiple of 4, got {num_frames}"
             )
 
         log.info(
@@ -498,7 +498,7 @@ def process(self, model, width, height, num_frames, hyvid_embeds, hyvid_embeds_2
             )
         if (num_frames - 1) % 4 != 0:
             raise ValueError(
-                f"`video_length-1` must be a multiple of 4, got {num_frames}"
+                 f"`video_length - 1 (that's minus one frame)` must be a multiple of 4, got {num_frames}"
             )
 
         log.info(